[
https://issues.apache.org/jira/browse/HDFS-4664?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13621269#comment-13621269
]
Steve Loughran commented on HDFS-4664:
--------------------------------------
Hadoop 1.3+ has the notion of "failure domains", as you can specify a topology
which declares that some nodes do not have independent failure, and that data
should not be replicated more than once on it.
Please look at HADOOP-8468 and see if that works -and if not, that should be
the starting point for changing things.
Assuming this JIRA is only about that, long-lived vs short lived DNs, how about
changing the title to something like "HDFS to support short-lived datanodes",
as that applies in more cases, such as in-cloud deployments with spot-priced
nodes
> HDFS for heterogeneous environment
> ----------------------------------
>
> Key: HDFS-4664
> URL: https://issues.apache.org/jira/browse/HDFS-4664
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: balancer
> Affects Versions: 0.20.2
> Environment: Ubuntu Linux, Institutional lab
> Reporter: Mohammad Mustaqeem
>
> I want to use HDFS for storing the files in the institutional labs.
> Here the point is to be noted that all the nodes in the labs are not of same
> type that is some nodes stay on for longer duration while some for small
> duration.
> In addition to this all the labs are not same means that some lab has UPS
> facility and some has more nodes.
> If I consider the lab as rack, then we should not choose the racks and node
> randomly in replica placement.
> We should give more priority to those nodes that stay on for longer duration
> and to those lab which has lab facility and has more system.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
